DWDRFeat: Describe Wider and Detect Robust Feature Extractor
Part Of
2024 IEEE 4th International Conference on Electronic Communications, Internet of Things and Big Data, ICEIB 2024
Journal Volume
33
Start Page
391
End Page
396
ISBN (of the container)
979-835036072-1
Date Issued
2024-04-19
Author(s)
Abstract
In the realm of computer vision, it is paramount to extract discriminative features, especially in dynamic settings characterized by shifts in illumination and viewpoint. Learnable feature extraction approaches have emerged as stalwarts in such scenarios, showcasing superior performance. However, their efficacy hinges on the acquisition of meticulously labeled, high-quality data, posing a substantial challenge. We delved into the intricate landscape of feature extraction and explored the potential of incorporating camera geometry constraints as valuable information. Leveraging corresponding points, more robust locations can be selected for feature points. Additionally, global information was augmented through attention mechanisms and wider receptive fields, enhancing the robustness of feature extraction in complex visual environments.
Event(s)
4th IEEE International Conference on Electronic Communications, Internet of Things and Big Data, ICEIB 2024
Publisher
IEEE
Type
conference paper